Predicting lock-induced waves in the intermediate channel of a composite navigation structure using a hybrid physics-informed and long short-term memory model

Composite navigation structures provide an effective solution for routing vessels past high dams, particularly in mountainous valleys. However, during lock discharge, the intermediate channel experiences complex wave–current interactions that significantly increase the difficulty of ship maneuvering. Using geometrically scaled physical model tests at a scale of 1:50, the dynamic evolution of lock-induced unsteady waves was systematically examined. Building on these insights, we develop a hybrid prediction model that integrates long short-term memory (LSTM) networks with physics-informed neural networks (PINNs) to achieve high-accuracy predictions of lock-induced wave processes. The waveforms in the intermediate channel are classified into two regimes by a dimensionless parameter Π: nonlinear Stokes-like waves for Π ≥ 5 and linear waves for Π < 5. The dominant frequency consistently falls within 0.2–0.4 Hz and is insensitive to the head difference, whereas higher heads concentrate wave energy and substantially increase wave amplitude. The proposed LSTM–PINN effectively captures nonlinear features in a water-level time series, and its physics-based constraints yield amplitude predictions in close agreement with experimental measurements. These findings clarify the classification mechanism and evolutionary behavior of wave–current structures in the intermediate channel. The hybrid LSTM–PINN framework offers a practical and accurate approach for predicting lock-induced waves, providing actionable guidance for the safe design and operational control of navigation facilities in high-dam projects.